Spiral bevel gears carry torque between intersecting or offset shafts in vehicle axles, helicopter drives, machine tools, and other high-speed transmissions. Their curved teeth allow gradual, oblique meshing, but the same geometry makes vibration behavior sensitive to small changes in the tooth surface. Contact pattern, loaded transmission error, and mesh force all shift when the cutter position or machine-tool setting changes. The design problem is therefore not simply to minimize a scalar mesh force, but to determine which directional component of the mesh excitation dominates the vibration response and which tooth-surface adjustment reduces it.
Directional vibration is the relevant design target
A gear mesh generates three force components and three moment components at the contact zone. Even when the resultant normal force is nearly unchanged, the distribution among these six components can change the vibration amplitude in a particular direction. A tooth-surface modification that reduces axial response may increase radial or angular response. For this reason, optimization based only on peak normal force, contact stress, or transmission error can miss the direction that actually dominates the housing or shaft vibration.
The optimization route connects four elements: a generated tooth surface, a loaded mesh-force calculation, a reduced finite-element model of the output shaft, and a six-degree-of-freedom sensitivity matrix. The tooth surface is controlled through machine-tool settings. The mesh calculation supplies the six force components. The output-shaft model supplies the harmonic response. The sensitivity matrix ranks the principal directions by compliance and provides weights for a scalar objective. The final step is a local adjustment of the machine settings within their admissible range.
A reduced output-shaft model for efficient harmonic response
A full spiral bevel gear transmission model normally assigns six degrees of freedom to each gear: three translations and three rotations. With two gears, the system has twelve degrees of freedom, and the equivalent mass, damping, and stiffness terms depend on the gear bodies, shafts, and bearings. Resolving the actual tooth contact inside such a model at every optimization iteration is expensive, especially when many candidate tooth-surface modifications must be compared.
The reduced approach keeps the output shaft as a discretized finite-element model and replaces the gear mesh with the six components of the mesh force. Gear, shaft, and bearing sections are represented by appropriate structural elements. The contact condition is no longer solved directly inside the finite-element model; instead, the mesh force calculated from tooth contact analysis and loaded tooth contact analysis is applied as external excitation. In the demonstrated case, this simplification changed the finite-element calculation from hours to seconds. The trade-off is that contact nonlinearity and local tooth compliance are carried by the force model rather than resolved explicitly at every node.
Tooth surface generation and six-component mesh force
The tooth surface of a spiral bevel gear is generated by a rotating cutter head. Cutter tip radius, profile angle, cutter tilt, and cutter swivel define the generating surface. Coordinate transformations from the cutter system to the generating gear and then to the work gear produce the theoretical tooth surface. Tooth contact analysis (TCA) locates the contact path and contact pattern; loaded tooth contact analysis (LTCA) adds the loaded deformation and determines the normal mesh force.
For vibration excitation, the normal force is projected into the local gear coordinate system as \(F_x\), \(F_y\), \(F_z\), \(M_x\), \(M_y\), and \(M_z\). The projection depends on the pressure angle, pitch cone angle, spiral angle, and the moment arms from the contact point to the relevant axes. In the optimization, the cradle swivel angle \(q_r\) and the machine root cone angle \(\Gamma_m\) are held fixed, while three settings are varied: horizontal wheel position \(X_r\), horizontal bed position \(X_{br}\), and vertical wheel position \(E_m\). These three parameters shift the generated surface and therefore change the contact condition and the six-component mesh force.
Six-degree-of-freedom sensitivity matrix and principal compliance
Let the generalized mesh-force vector and the generalized displacement vector be
\[
\mathbf F=[F_x,F_y,F_z,M_x,M_y,M_z]^{T},\qquad
\mathbf u=[\delta_x,\delta_y,\delta_z,\theta_x,\theta_y,\theta_z]^{T}.
\]
Under linear elastic, small-deformation conditions, the displacement produced by a unit generalized force defines the compliance matrix
\[
C_{ij}=\frac{\delta_i}{F_j},\qquad C\in\mathbb{R}^{6\times 6}.
\]
For a stable elastic system, \(C\) is symmetric and positive definite. Its eigenpairs satisfy
\[
C\mathbf v_i=\lambda_i\mathbf v_i,\qquad
\lambda_1\ge\lambda_2\ge\cdots\ge\lambda_6>0.
\]
The eigenvalue \(\lambda_i\) is the principal compliance in the direction of eigenvector \(\mathbf v_i\). A large eigenvalue means that a unit generalized force along that principal direction produces a relatively large displacement; the system is flexible in that direction. A small eigenvalue indicates a stiffer direction. The eigenvalues therefore form a directional stiffness fingerprint of the gear-shaft system at the mesh location.
To convert the six-objective problem into a scalar objective, the eigenvalues are normalized:
\[
\omega_i(\mathbf x)=\frac{\lambda_i(\mathbf x)}{\sum_{j=1}^{6}\lambda_j(\mathbf x)},\qquad
\sum_{i=1}^{6}\omega_i=1,
\]
where \(\mathbf x\) denotes the vector of adjustable machine settings and \(C(\mathbf x)\) is the resulting compliance matrix. The weighted objective is
\[
\min_{\mathbf x} U(\mathbf x)=\sum_{i=1}^{6}\omega_i(\mathbf x)\lambda_i(\mathbf x).
\]
The weight \(\omega_i\) measures the share of the total principal compliance carried by the \(i\)-th principal direction. A direction with a large weight contributes strongly to the overall flexibility and is a priority for adjustment. In the case examined, the \(x\)-direction response received the largest weight, so the optimization focused on reducing that component while keeping the other directions under observation.
Case configuration and machining settings
The demonstration used a 28-tooth pinion and a 41-tooth gear with a 90° shaft angle. The transverse module was 3.7500 mm, the pressure angle was 20°, and the spiral angle was 18.3000°. The outer cone distance was 93.0914 mm. Both members and shafts were represented with 45 steel properties, using a Poisson ratio of 0.3 and a density of 7.86 g/cm³. The load case was 400 N·m.
Table 1. Principal geometry of the spiral bevel gear pair.
| Parameter | Pinion | Gear |
|---|---|---|
| Number of teeth | 28 | 41 |
| Transverse module (mm) | 3.7500 | 3.7500 |
| Pressure angle (°) | 20 | 20 |
| Spiral angle (°) | 18.3000 | 18.3000 |
| Outer cone distance (mm) | 93.0914 | 93.0914 |
| Addendum (mm) | 3.9675 | 2.4075 |
| Dedendum (mm) | 3.1125 | 4.6725 |
| Tip angle (°) | 2.8734 | 1.9150 |
| Root angle (°) | 1.9150 | 2.8734 |
| Pitch diameter (mm) | 105 | 153.75 |
| Pitch cone angle (°) | 34.3302 | 55.6698 |
The cradle swivel angle and machine root cone angle were fixed. Only the three translational machine settings were adjusted, which keeps the optimization local and avoids changing the basic cutter geometry or tooth count.
Table 2. Machine-tool settings before and after optimization.
| Machining setting | Initial (mm) | Adjusted (mm) | Change (mm) |
|---|---|---|---|
| Horizontal wheel position, \(X_r\) | 18.30 | 18.04 | −0.26 |
| Horizontal bed position, \(X_{br}\) | 34.30 | 34.33 | +0.03 |
| Vertical wheel position, \(E_m\) | 18.30 | 17.00 | −1.30 |
The adjustment is highly local. The horizontal bed position changed by only 0.03 mm, and the horizontal wheel position changed by 0.26 mm. The vertical wheel position changed by 1.30 mm, which is the largest shift and therefore the main geometric lever in this case.
Result: a 20.21% reduction in the dominant x-direction response
At 400 N·m, the initial tooth surface produced a mesh-force amplitude of 8,034 N. The corresponding signed \(x\)-direction response amplitude at the 1 Hz comparison point was \(-5.7637\times10^{-3}\). After the three machine settings were adjusted, the mesh-force amplitude was 7,995 N and the \(x\)-direction response amplitude was \(-4.5986\times10^{-3}\). Comparing magnitudes, the dominant \(x\)-direction response fell by 20.21%.
The weighting result followed the same direction. The dominant weight \(\omega_1\) fell from 0.70 to 0.45, a relative reduction of 35.7%. The \(x\)-direction still carried the largest share after optimization, but its share of the total principal compliance was substantially lower. This is consistent with a local stiffness improvement in the dominant direction rather than a complete change of the system’s vibration mode.
Table 3. Mesh force, weighting, and \(x\)-direction response before and after adjustment.
| Metric | Initial | Adjusted | Relative change |
|---|---|---|---|
| Mesh-force amplitude (N) | 8,034 | 7,995 | −0.49% |
| Dominant weight \(\omega_1\) | 0.70 | 0.45 | −35.7% |
| \(x\)-response magnitude (\(\times10^{-3}\)) | 5.7637 | 4.5986 | −20.21% |
The scalar mesh-force amplitude changed by less than 0.5%, while the \(x\)-direction response changed by more than 20%. The improvement therefore cannot be explained by a large reduction in the total mesh force. It comes mainly from a change in the six-component force distribution and the directional compliance seen by the dominant response. The tooth surface acts as a directional filter: a small geometric shift redistributes the excitation among translation and moment components, and the sensitivity weight tracks that redistribution.
What the result establishes and where it stops
The finite-element case supports the internal consistency of the workflow. The sensitivity matrix identifies the \(x\)-direction as dominant, the machine-setting adjustment reduces that direction, and the normalized weight falls accordingly. The result also shows that machine-tool settings can be used as a low-dimensional control space for local tooth-surface optimization without rebuilding the entire transmission for every trial.
The evidence is numerical rather than acoustic. It does not establish a broadband noise reduction, a fatigue-life improvement, or robust performance over the full speed and load map. The model assumes linear elastic behavior, uses a single torque level and a single comparison frequency, and relies on the TCA/LTCA force model for contact behavior. Manufacturing tolerances, thermal distortion, lubrication, and contact-pattern robustness require separate checks. The 20.21% reduction is therefore a conditional result for the stated geometry, load, and response direction, not a general guarantee for all spiral bevel gear pairs.
The practical value of the method is as a screening and local-tuning tool. It answers a specific design question: given a generated tooth surface and a dominant vibration direction, which machine setting should be changed, and how much does the directional compliance improve? That is a narrower claim than full transmission validation, but it is a useful one because it links tooth-surface geometry, six-component mesh excitation, and directional vibration in a single optimization loop.